Using Three-Cup Shuffle Magic Trick as a backdrop, we present a system which performs head, body pose, and fiducial-based object tracking all in 3D as well as hand and head gesture recognition and production. The sensor born information and the capabilities are tightly integrated in a psychologically plausible manner within an embodied version of the ACT-R cognitive architecture, ACT-R/E. The Mobile-Dextrous-Social (MDS) Robot interprets and uses gestures, including deictic and symbolic hand gestures as well as head nods and shakes. The overall approach provides us with a powerful, integrated system facilitating many different avenues of research in human-robot interaction.